src2pkg-2.7 - create packages from any source code

[amigo]

Yes, thank you. I'll see about morphing that into a puppy-specific extension for src2pkg and then everyone can be happy with more complete *.specs files.

[Keef]

I've just tried to install the pet on Akita (beta 7), but it fails when trying to compile libsentry. Tried running make on libsentry (in the src2pkg-helpers dir), and got the following result:

Code: Select all

# make
cc -Wall -U_FORTIFY_SOURCE -D_GNU_SOURCE -DLIBDIR=\"/usr/local/lib\" -DPIC -fPIC -D_REENTRANT -DVERSION=\"0.7.0\" -c libsentry.c
libsentry.c:3090: error: conflicting types for 'scandir'
/usr/include/dirent.h:254: error: previous declaration of 'scandir' was here
libsentry.c:3743: error: conflicting types for 'scandir64'
/usr/include/dirent.h:275: error: previous declaration of 'scandir64' was here
make: *** [libsentry.o] Error 1

It has worked ok for me on Fire Hydrant Lite 301 RetroV3, and SolidPup (Wary 511).

[amigo]

Ay, Ay, Ay, that old problem -it's a glibc thing, where it is declared differently in different versions.
What version of gcc do you have ( output from running: gcc -dumpversion)?
What version of glibc do you have (output from running: /lib/libc.so.6)?

[Keef]

Amigo,

Here are the gory details:

gcc version is 4.2.2

Code: Select all

# /lib/libc.so.6
GNU C Library stable release version 2.10.1, by Roland McGrath et al.
Copyright (C) 2009 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.
There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A
PARTICULAR PURPOSE.
Compiled by GNU CC version 4.3.4.
Compiled on a Linux >>2.6.32.16<< system on 2010-09-13.
Available extensions:
        crypt add-on version 2.1 by Michael Glad and others
        Native POSIX Threads Library by Ulrich Drepper et al
        Support for some architectures added on, not maintained in glibc core.
        BIND-8.2.3-T5B

EDIT I've had a PM from scottman about this, with a possible solution. Will update when I've tried it.

[Tman]

Hi Amigo,

When src2pkg creates a pet from a source on my hard-drive, the pet contains the proper tree structure to files and folders but it contains a secondary tree where all of the files are duplicated inside of a folder named initrd. I was wondering if there is a way to avoid this?

[Tman]

removed double post

[amigo]

Keef, Notice this:
gcc version is 4.2.2
and this?:
Compiled by GNU CC version 4.3.4
May be part of the problem. There is a small patch that is used in the libsentry sources which needs to be removed or applied -mostly depending on the glibc version. It comes applied with the latest versions of src2pkg, but maybe I need more 'ifs' in the code somewhwere to try to get it right for everybody. Your feedback is useful. But that version mis-match you have there is not a good sign... Puppy is infamous for its' wacky, unclean toolchains...

[amigo]

Tman, what software are you trying to build. I have never seen a report of this problem for Puppy -it's a very good sign that at least examine the contents of a potential package.

The specific software you are building may have something to do with this, but I may have to build in a special work-around for this situation.
Please cd into the package dir where the content it and run 'find > files.txt'
The cd into your /initrd dir and do the same, then post the results.

If you are running from hard drive, why is there still an /initrd dir around anyway?

[Tman]

I don't think that the reason is becuase of the sources. When I last compiled Thunar for dpup Exprimo, Thunar and all of the dependency pets contained 2 trees.

I don't know if it has anything to do with how I set up my hard-drives.
My first hard-drive has Windows XP and a secondary Linux partition which I plan to install debian onto.

The second hard-drive has a Windows7 partition, another ntfs partition for storage, and the third ext4 partition for frugal pups.

I download the sources to my /mnt/home (2nd hard-drive), which is an ext4 partition.

The reason why I download the sources first is because I extract them and run ./configure --help to see all of the options. I next type in all of the options that I want into a text file, which I later paste into LXTerminal when running src2pkg.

[Tman]

Please cd into the package dir where the content it and run 'find > files.txt'

I quickly recompiled jwm which I tried before, but deleted after.

Code: Select all

cd /mnt/home
src2pkg -VV jwm-2.1.0.tar.bz2

Below is the contents of the files.txt with the contents of the normal tree and the initrd tree.

Code: Select all

.
./files.txt
./jwm-2.1.0-i486-1
./jwm-2.1.0-i486-1/initrd
./jwm-2.1.0-i486-1/initrd/mnt
./jwm-2.1.0-i486-1/initrd/mnt/dev_save
./jwm-2.1.0-i486-1/initrd/mnt/dev_save/jwm-2.1.0-src-1
./jwm-2.1.0-i486-1/initrd/mnt/dev_save/jwm-2.1.0-src-1/src
./jwm-2.1.0-i486-1/initrd/mnt/dev_save/jwm-2.1.0-src-1/src/jwm
./jwm-2.1.0-i486-1/usr
./jwm-2.1.0-i486-1/usr/etc
./jwm-2.1.0-i486-1/usr/etc/system.jwmrc
./jwm-2.1.0-i486-1/usr/share
./jwm-2.1.0-i486-1/usr/share/man
./jwm-2.1.0-i486-1/usr/share/man/man1
./jwm-2.1.0-i486-1/usr/share/man/man1/jwm.1
./jwm-2.1.0-i486-1/usr/man
./jwm-2.1.0-i486-1/usr/man/man1
./jwm-2.1.0-i486-1/usr/man/man1/jwm.1.gz
./jwm-2.1.0-i486-1/usr/doc
./jwm-2.1.0-i486-1/usr/doc/jwm-2.1.0
./jwm-2.1.0-i486-1/usr/doc/jwm-2.1.0/jwm.desc
./jwm-2.1.0-i486-1/usr/doc/jwm-2.1.0/LICENSE
./jwm-2.1.0-i486-1/usr/doc/jwm-2.1.0/README
./jwm-2.1.0-i486-1/usr/bin
./jwm-2.1.0-i486-1/usr/bin/jwm

??. I guess I was wrong about the initrd tree duplicating everything. In this instance, it only contains the jwm binary.

[Tman]

Btw.. I didn't attach the file.txt for the Thunar and its dependency pets becuase I altered them and deleted the original pets.

[amigo]

dev_save -there's the problem. I'm afraid I'll have to look at a fix for this tomorrow as it is getting late here. It would help if you post the output from scr2pkg when you run this build. What I'm looking for is what comes just after this line in the output:
Compiling has been - Successful!"
Make sure you are not using the '-VV' option or the line will be lost.

No wait, here's a fix for it. Open the file /usr/libexec/src2pkg/08-fake_install and scroll down to line 303:
"post_fake_install() {"
Place the cursor at the end of that line, hit ENTER and then paste this in there:

Code: Select all

	if [[ -d "$PKG_DIR"/initrd ]] ; then
		( cd "$PKG_DIR" && rm -rf initrd )
	fi

There's already a general fix-up routine there which covers all content-creation types, so it should be fine there.

[Tman]

Thanks, very much, amigo.
Your patch fixed the issue in the pet. It is still creating the initrd folder, but your patch removes it.

I here is the output from src2pkg as you requested earlier:

Code: Select all

Compiling has been - Successful!
Checking for Makefile rule: 'install' Okay
Checking support for DESTDIR (or similar) - Not found
   Notice - DESTDIR support not found. Falling back to JAIL install
Creating content in JAIL root - Using: 'make install'
cd src ; make all
make[1]: Entering directory `/initrd/mnt/dev_save/jwm-2.1.0-src-1/src'
make[1]: Nothing to be done for `all'.
make[1]: Leaving directory `/initrd/mnt/dev_save/jwm-2.1.0-src-1/src'
cd src ; make install ; cd ..
make[1]: Entering directory `/initrd/mnt/dev_save/jwm-2.1.0-src-1/src'
strip jwm
install -d /usr/bin
install jwm /usr/bin/jwm
make[1]: Leaving directory `/initrd/mnt/dev_save/jwm-2.1.0-src-1/src'
install -d /usr/etc
install -m 644 example.jwmrc /usr/etc/system.jwmrc
install -d /usr/share/man/man1
install -m 644 jwm.1 /usr/share/man/man1/jwm.1
   Notice - Checking integrity of '' installation - Done
Safe content creation - Successful!
Processing package content:
Correcting package permissions - Done
Stripping ELF binaries - Using: strip -p --strip-unneeded Done
Checking for standard documents - Done
   NOTICE - No man-pages installed by Makefile
Found man-pages in SRC_DIR - Installing in PKG_DIR
Compressing man pages - Done
Creating slack-desc - From default text
Rechecking package correctness:
Checking for misplaced dirs - Done
Rechecking package permissions - Done
Checking for dead links in PKG_DIR - Done
Searching for links in: jwm-2.1.0-i486-1 - None found
Creating package: 'jwm-2.1.0-i486-1.pet' - Done
Package Creation - Successful! Package location:
/mnt/home/jwm-2.1.0-i486-1.pet

[amigo]

"Creating content in JAIL root" is what I was looking for -as I thought I'd have to code something for each type of content-creation.

"It is still creating the initrd folder" -I'll fix this a different way for next version which should avoid it being created at all.

[amigo]

Keef, I need to see a copy of a header on your system:
/usr/include/dirent.h
Either paste inside code tags here, or attach it please. I'm pretty sure there is a mis-match between that header and the version used when glibc was compiled. When src2pkg tries to compile libsentry, the libsentry 'make' process tries to figure out which syntax to use for the 'scandir' function among others. The glibc version is read directly to figure this out. A new syntax is used for glibc >= 2.10, so it should be working. But that syntax should also match in the 'kernel headers' which were used to define the glibc kernel interface when glibc was compiled. If they don't compiling *lots* of things will fail.

I'm pretty sure that I can't really fix this in the libsentry make process. If I have it read the scandir syntax from the header, instead of from implying from the glibc version, I think the library would compile, but then segfault when you run it. Could you perhaps try some other distro with a 'sane' toolchain? No, I didn't really mean that did I?

[sc0ttman]

Keef, I need to see a copy of a header on your system:
/usr/include/dirent.h....I'm pretty sure there is a mis-match between that header and the version used when glibc was compiled

Just in case you're gonna waste your own precious time on this, Keef and myself are past this now, all fixed...
It was indeed me being dumb/lazy/cautious... I got my arse in gear, the toolchain is now 'sane'
(ish... its still got old kernel, therefore old kernel headers....)

... But I just built libsentry fine, and went on to compile lots of other stuff, inc FreeCiv, some SDL games, etc...

I think the library would compile, but then segfault when you run it.

Leave as is, so others hit the same problem, if they deserve it!

[amigo]

Gosh, scOttman, I feel so... vindicated! You must have wrongly updatet the kernel headers when you updated the kernel.... For my distro, I call the headers 'libc-headers' instead of kernel-headers so that no one will think they are really related to the running kernel version. No ,wait, I missed this: "(ish... its still got old kernel, therefore old kernel headers....)".

Let me reiterate. When you compile glibc for a new system, you choose a certain version of the includes from the kernel sources. The version of these headers has absolutely no relation to whatever kernel you are running at glibc compile time, nor any relation to the kernel version you plane to release, nor to any upgraded-kernel you might use afterward (well, there a couple of corner cases where this is not *absolutely* true).

The headers used when compiling glibc define how glibc interfaces with the kernel. In order to have a sane system, these same kernel-headers should be used for compiling all software on the system until a new glibc version is created. So, the kernel headers which are present under /usr/include should always be the same for the life of the distro. And note that as soon as you upgrade glibc, you should properly re-compile *every* bit of software on the system. If you have noticed the many GLIBC_VERSION errors on this forum, this is exactly where they come from -mismatching binaries/libs.

This does not mean that you can never upgrade the kernel. The kernel sources do not install their headers under /usr/include -the kernel includes everything it needs under one dir. So, go right ahead and run any kernel version you like. The only time that the headers of the running kernel version matter is when you are compiling out-of-tree kernel modules.

So, make a nice package out of the original headers so they can always be handy and stay the same. Then, do what you like about a running kernel. When you need to compile out-of-tree modules, they are gonna need the full kernel sources *of the running kernel*. Compiling the module will look for the internal kernel headers -usually in the default location /usr/src/linux -otherwise there will be a configuration option which points the build at the correct kernel version.

Read this over and over again until it starts to make sense.

Note that there are other elements to having a truly sane toolchain. The toolchain only consists of three things: glibc, gcc and binutils. Later versions of gcc have external lib dependencie which also should fit and match, but to really understand the basics just consider those three.

Now, a truly sane toolchain is when all three of these components have a 'circular dependency* on each other. That means that *this* glibc was compiled by *this* gcc and linked using *this* binutils. It takes at least two passes to accomplish that. If that sounds like overkill, then you'll never be up to creating your own toolchain. Every major distro -nay, every 'true' distro does this every time the toolchain changes. Why? It is the Right Way (tm) and the only way to ensure that you never see those GLIBC_VERSION errors. Any so-called distro which does not observe these procedures is pure junk and is gonna bring lots of misery to nearly everyone who touches it. You can't simply grab some binaries from just anywhere and start mixing them.

[Keef]

Amigo,
Thanks for the explanation. It is still mostly above my skill level, but I think I get the gist of it. I can understand the need to do several passes when building new tools. There is probably a good analogy somewhere, but I can't think of it yet. If you are using the 'old' tools to build the 'new', the job's not done until you have 'fully-filtered' new tools that can completely replace all the 'old' ones.

[amigo]

Yeah, you build the new tools with the old ones -upgrading as you go, then re-build it all using the new versions -done.

[c.lei]

Not really puppy related, but I think everyone should be aware of this:

For some reason unknown, installing src2pkg-2.7-2.noarch.rpm on my machine running Fedora 9 corrupts the system.

When I reboot the system, I get a Gnome system error message basically saying the Gnome Power Management defaults have been corrupted and the
Gnome display is pretty much trashed.

Basically I have to reinstall the system from a recent partition image backup to recover from this since I really don't have a clue as to what acutally got trashed, and I'm not about to install the rpm on my other computers.

Any ideas as to what is going on other that a bad install rpm that's somehow trashing non-related files in ect when it's installed on Fedora and other systems?